Today's centralization, when did the open internet become Cloudflare, AWS, and Google?

Field note. Three companies sit between your traffic and most of your players. That's a lot of leverage in few hands, and it shapes what 'reliable hosting' even means.

The early internet was a decentralized vision. Many small networks, many independent operators, no central authority. The end-to-end principle. Permissionless innovation.

The internet of 2026 is genuinely more centralized than the internet of 1996. A handful of companies host most of the world's services. A handful of CDNs front most of the world's websites. A handful of platforms mediate most communication.

This is a long-running tension. This article surveys where we actually are and how we got here.

What centralization means here

The internet's centralization happens at several layers, and they're worth separating:

Infrastructure (where things are hosted). AWS, Microsoft Azure, Google Cloud, Alibaba Cloud. Together, ~60-70 percent of cloud workloads.

CDN / network edge (what sits in front of services). Cloudflare, Akamai, Fastly, AWS CloudFront. Cloudflare alone protects ~20 percent of websites globally.

DNS resolution. Public resolvers from Google (8.8.8.8), Cloudflare (1.1.1.1), and a few others handle a huge fraction of DNS queries.

Search. Google has ~90 percent global search market share. Bing and others fill the rest.

Social and communication. Meta (Facebook, Instagram, WhatsApp), Google (YouTube, Gmail), Microsoft (LinkedIn, Teams), X, TikTok.

Mobile platforms. Apple iOS and Google Android. Effectively a duopoly.

Browsers. Chromium-based browsers (Chrome, Edge, Brave, etc.) plus Safari ~95 percent of the market.

The internet's infrastructure is more centralized than its participation. Anyone can publish a website. Most websites are hosted on, served by, or accessed through services owned by a handful of companies.

How we got here

This wasn't inevitable. Several specific dynamics:

Economies of scale. Large infrastructure providers offer better service at lower cost. Small competitors lose on both axes.

Network effects. Some services (social networks, marketplaces) get more valuable as they get larger. Big players grow naturally.

Acquisition strategies. Big companies bought competitors before they became threats. WhatsApp and Instagram by Facebook. YouTube and DoubleClick by Google. LinkedIn and GitHub by Microsoft.

Platform leverage. Apple and Google use their mobile-OS positions to favor their own services and disadvantage rivals.

Regulatory inaction. For most of the 2000s and 2010s, antitrust enforcement in tech was minimal. Companies that should have been broken up weren't. Mergers that should have been blocked weren't.

Cloud convenience. Building on AWS / Azure / GCP is easier than building on alternatives. Network effects on infrastructure.

Privacy regulations had unintended centralizing effects. GDPR compliance is easier for big companies with legal teams; small companies struggle and exit.

The combined effect: each year, more of the internet's value flowed to fewer companies.

Specific concentration metrics

A few worth knowing:

Cloud market share (2026 approximate):

• AWS: ~30 percent.
• Azure: ~25 percent.
• Google Cloud: ~12 percent.
• Alibaba Cloud: ~5 percent.
• Others: rest.

The top three combined: ~67 percent.

CDN market share:

• Cloudflare: ~20-25 percent of websites.
• Akamai: large in enterprise.
• AWS CloudFront, Fastly, others.

Search:

• Google: ~90 percent globally.
• Bing: ~3 percent.
• Yandex (Russia, neighboring), Baidu (China): regional dominance.

Mobile OS:

• Android: ~70 percent.
• iOS: ~30 percent.

Browser engines:

• Chromium (Blink): ~70 percent.
• WebKit (Safari): ~20 percent.
• Gecko (Firefox): ~3 percent.

When 3 engines run most of the web, the engine developers (Google, Apple, Mozilla) effectively shape what the web can do.

The costs

Concentration has real costs:

Outages. When AWS US-EAST-1 has problems, large slices of the internet go down. When Cloudflare has a routing issue, thousands of sites become unreachable. The 2021 Facebook outage, the 2017 S3 outage, the 2022 Cloudflare outage, all demonstrated.

Surveillance capacity. Centralized services see enormous quantities of user activity. Governments that pressure these services get pervasive surveillance reach.

Censorship vulnerability. A few platforms decide what speech is allowed at scale. Their decisions affect billions.

Innovation friction. Competing with a hyperscaler is expensive. Startups in some categories face high entry costs.

Pricing power. Concentration enables price increases. Some cloud prices have risen even as costs have fallen.

Single points of dependency. Many businesses depend on AWS, Stripe, Cloudflare, Google. Any disruption ripples widely.

The benefits

Centralization also produced real benefits, which is why it happened:

Operational excellence. Hyperscalers run better infrastructure than most smaller operators could. Reliability, security, performance: usually better.

Cost reduction (for many). For small users, hosting on AWS is cheaper than building your own data center.

Innovation enabled. Building on existing platforms lets small teams ship products that previously required infrastructure investments.

Standards leadership. Big providers push web standards forward (HTTP/3, modern crypto, etc.).

Talent concentration. Big companies attract talent that delivers value.

Centralization isn't pure loss. It's a trade.

The reactions

Multiple counter-trends have emerged:

Decentralized protocols. ActivityPub (Mastodon, Threads), ATproto (Bluesky), Matrix (messaging). Federated alternatives to centralized social networks.

Open-source clouds. Open-source projects (Kubernetes, OpenTelemetry, etc.) reduce lock-in. "Cloud-native" but portable.

Repatriation. Major customers moving workloads off cloud back to dedicated infrastructure (covered in the cloud era article).

Regulation. EU's Digital Markets Act (2022), various antitrust cases (US v. Google, EU v. Meta), GDPR (2018). Slow, often imperfect, but accumulating.

Self-hosting movement. Small communities running their own services (Mastodon instances, ATproto PDSes, etc.). Niche but growing.

Indie web. Personal websites, RSS, modest hosting. A reaction to platform dominance.

None of these alone reverses centralization. Together they slow it and provide alternatives.

Specific dramas

Some specific events of the 2010s and 2020s:

Net neutrality fights. Whether ISPs can prioritize traffic from specific providers. U.S. policy has flipped multiple times. Concentrated providers have favored some outcomes.

The platform speech debates. Twitter / X, Facebook, YouTube deciding what speech to allow. Each major decision attracts political and legal attention.

The Apple-Epic case (2021). Whether Apple's 30 percent App Store fee is an antitrust violation. Mixed ruling.

The Microsoft-Activision case (2023). Whether the merger should be allowed. After regulatory drama, the merger went through.

The Meta-WhatsApp delayed integration. Regulators required WhatsApp to stay somewhat separate from Facebook for years.

Cloudflare's content decisions. Cloudflare has occasionally cut off services it considers ethically beyond the pale (Daily Stormer 2017, Kiwi Farms 2022). Each decision raises "who decides" questions for centralized infrastructure.

The TikTok bans. Concerns about Chinese ownership of a major platform. Multiple countries have restricted or banned TikTok.

Browser engine dependency. Apple's WebKit-only requirement on iOS browsers eased under EU pressure in 2024. The implications continue to be debated.

The 2020s have been characterized by visible friction between centralized platforms and various regulators, users, and competitors.

Where the gaming sector sits

Gaming has its own concentration story:

Console platforms. Sony PlayStation, Microsoft Xbox, Nintendo. Three companies control most console gaming.

PC distribution. Steam dominates. Epic and others minor.

Mobile gaming distribution. App Store and Google Play. Two channels.

Game engines. Unity and Unreal dominate. A small open-source scene (Godot) exists.

Online services. PlayStation Network, Xbox Live, Steam, Epic, Nintendo Online. Each is locked to its platform.

For game hosting, the concentration is somewhat less extreme. Many small hosting providers serve specific games or communities. The "gaming hosting" market is more fragmented than general cloud hosting. This is partly because games have specific requirements (latency, custom protocols) that hyperscalers don't always serve as well as specialists.

What's coming

A few trends to watch:

Multipolar cloud. Alibaba, regional providers, sovereign clouds. China's tech sphere is largely separate from the U.S. one. Other regions may follow.

Regulatory action. EU is most aggressive; U.S. and others slowly catching up. Real antitrust actions may continue.

Decentralized protocols slowly maturing. ATproto, Matrix, ActivityPub each grow. Their network effects compound over years.

Self-hosting renaissance. Easier tools (Tailscale, Cloudflare Tunnel, simple Linux distros) lower the barrier. Niche but real.

The Web platform vs apps. Browsers and apps will continue contesting which is the right way to build software. Apps win on some axes; the Web on others.

AI implications. Large AI models are themselves enormously concentrated. The pattern could deepen or shift.

A balanced view

Centralization isn't pure loss. The hyperscalers genuinely deliver real value: better performance, better security, lower costs (for many), faster innovation. Throwing all this away in the name of decentralization would be a bad trade.

What's worth fighting for: the option to leave. The ability to migrate. The presence of alternatives. The transparency and accountability of dominant providers. Realistic competition.

These can coexist with centralization-as-a-feature. The internet of 2030 will likely still have hyperscalers, still have CDN concentration, still have platform giants. The question is whether the surrounding structure (alternatives, regulations, norms) preserves the original openness as a meaningful possibility.

The original internet wasn't perfect

A note for perspective: the early internet's "decentralization" was partly an artifact of the technology being immature. Many users had no good options for hosting; running your own server was hard. The flattening we now see also reflects services genuinely working better at scale.

There's no era to "go back to." There's only the next era, which will be shaped by today's choices.

Conclusion

The internet is more centralized than it was, in specific ways. This is partly the natural result of useful scale, partly the result of business strategies, partly the result of regulatory neglect.

The costs are real (outages, surveillance reach, single points of failure). The benefits are also real (operational quality, accessibility for small users).

The future depends on whether decentralization-supporting protocols mature, regulators act effectively, and users care enough to choose alternatives where they exist.

This is the open question of the internet's next decade. Watch it with knowledge of the long arc behind it.

End of Series C

The 10-article internet history series is complete:

1. ARPANET.
2. TCP/IP's triumph.
3. The DNS story.
4. Tim Berners-Lee and the Web.
5. The browser wars.
6. The dot-com era.
7. Mobile internet.
8. The cloud era.
9. IPv4 exhaustion.
10. Today's centralization.

The internet has had 55 years of evolution. Most of what it does now would have astonished its founders. Some of what it has become would have disappointed them. Both are true.

The next series turns to computing history more broadly: the inventions of operating systems, languages, and software movements that made the internet's transformation possible.